Problem

With Tasks with tags and a lot of characters that are specific to the json-l format (like parantheses and quotation marks) LLMs tend to break the syntax of the json-l

Solution ideas

a) JSON:
Since we can not gather Details, what exactly failed, we switch to single-segment processing as before.
Batch-Logging should be enriched with task name/id and resource name /id

b) XLIFF
Here we re-request the failed Batch starting from the broken Segment and specify, what went wrong (-> the tag that was missing or had a broken structure). Currently, we do not switch to single-segment processing here but log an error when the batch fails after the second attempt. If it turns out, that this happens too often, we have to enable single-segment processing and add tag-repair to the XLIFF handler - OR make the handler changable for single-segment processing